Wang Zi

CASE 05 / ENGINEERING IN DETAIL

Dify Self-Service Assistant for Operations

Let operations query internal systems from 360Teams by translating questions into API calls.

Qifu Technology (奇富科技) · Project contributor · 2025.05 — 2026.03

Context and constraints

Operations staff regularly asked engineering to look up data, statuses, and business rules. I contributed to a project that used Dify to recognize intent, extract parameters, call existing internal APIs, and return results through a 360Teams bot. This case describes the project workflow; my role was as a contributor.

Core workflow

This diagram is a simplified view of the technical approach.

  1. 01 / Operations question360Teams bot
  2. 02 / Question understandingIntent recognition and parameter extraction
  3. 03 / Business queryCalls to existing internal APIs
  4. 04 / Response assemblyAnswers based on query results

Design decisions and implementation

01

Inspect intent, parameters, and API calls separately

ChallengeOperations staff describe a business question; internal APIs require a specific operation and explicit parameters.

ApproachThe project used Dify to orchestrate intent recognition, parameter extraction, internal API calls, and response assembly as separate steps.

ReasoningWhen a query fails, the intent and parameters can be checked before examining the API response.

02

Build replies from internal API results

ChallengeData, statuses, and rules each have their own sources. Language understanding must connect to those business facts.

ApproachWorkflows pass extracted business parameters to existing internal APIs and compose replies from the returned results, covering routine operations queries.

ReasoningBusiness data in a reply has a corresponding API response, making it possible to check for misread or omitted values.

03

Provide the query entry point in 360Teams

ChallengeEven a usable query tool adds friction if staff must search across systems to find the right entry point.

ApproachThe project integrated a 360Teams bot so operations staff could ask questions directly, bringing data lookups, status checks, and rule queries into their everyday communication tool.

ReasoningOperations staff can initiate these queries in their existing communication tool without finding a separate system entry point.

What the design addresses

The project combined intent recognition, parameter extraction, and internal API calls in Dify workflows for data, status, and rule queries through 360Teams.

Validation focus

  • Can the workflow identify information that needs clarification when a question or its parameters are incomplete?
  • Are API permissions, error responses, and empty results included in pre-release acceptance checks?
  • Does the answer agree with the business API response, and can that source be checked?

My contribution

Contributed to the operations self-service query project, working with the team on routine data, status, and business-rule lookup scenarios.

Technologies and topics

  • Dify
  • Intent recognition
  • Parameter extraction
  • Business APIs
  • Workflow orchestration
  • 360Teams

Let’s talk about what you’re building.

Share your goals, current systems, and delivery scope. We can work through the key technical questions and a practical path forward.

Send an email ↗